]> Continued Fractions - Matrices

Matrices

We'll approach continued fractions from a completely different viewpoint.

Let A 1 ,... be a sequence of 2 ×2 matrices. Let

(p n r n q n s n)=A 1 ...A n

If p nq n and r ns n tend to the same limit α as n then α is the limit of the infinite product of matrices, and we write

α~ n=1 A n

Example: Let a 1 ,... be a sequence of positive integers, except we allow a 1 =0 . Let

A k=(a k 1 1 0 )

Then [a 1 ;a 2 ,...]~ n=0 A n. The convergents are p n,q n, and we have r n=p n1 ,s n=q n1 . Since detA n=1 , we see the difference between successive convergents can be determined by studying

detp n r n q n s n=(1 ) n

Example: Suppose α~ k=1 A k.

Let c k be nonzero numbers. Then α~ k=1 c kA k.

Let A=(a b c d). Then

aα+bcα+d~A k=1 A k

provided cα+d0 . For some proofs, we exploit this and ignore a finite number of matrices.

Example: Given that (1 2 1 1 )(1 2 1 1 )... converges, find its limit.

Let

α~ n=1 (1 2 1 1 )

Then

(1 2 1 1 ) n=2 (1 2 1 1 )~α+2 α+1

Hence α=α+2 α+1 , and therefore α=2 .

Theorem: Consider a sequence of matrices of the form A k=(a k c k b k 0 ). where a k,b k,c k are positive integers satisfying

a k(b kc k) 1 +δ

for some fixed δ>0 . Then n=1 ~α for some irrational α>1 .

Proof: Let k=1 nA k=(p n r n q n s n). We have r n=c np n1 ,s n=c nq n1 , q as n and p nq n. The last fact implies α>1 .

Also,

p n+1 q n+1 =a n+1 p n+b n+1 c np n1 a n+1 q n+b n+1 c nq n1

so p n+1 q n+1 lies strictly between p nq n and p n1 q n1 .

Then:

p nq np n1 q n1 = k=1 ndetA kc nq nq n1 =b 1 c 1 ...b nc nc nq nq n1 <b 1 c 1 ...b n1 c n1 (q n1 ) 2

since q n>b nq n1

Now q n1 >(kb 1 c 1 ...b n1 c n1 ) 1 +δ for some κ>0 because

q n1 = a n1 q n2 +b n1 c n2 q n3 a n1 q n2 (b n1 c n1 ) 1 +δq n2

and if we asssume inductively that q n2 >(κb 1 c 1 ...b n2 c n2 ) 1 +δ it follows that

q n1 >(κb 1 c 1 ...b n1 c n1 ) 1 +δ

So p nq np n1 q n1 0 as n and

αp n1 q n1 p nq np n1 q n1 <b 1 c 1 ...b n1 c n1 q n1 2 <1 κq n1 1 +σ

where σ>0 is defined by 1 1 +δ=1 σ.

Hence α is irrational, because if α=k/l then:

kq nlp n=lq nαp nq n<lq nκq n 1 +σ0

as n, a contradiction because kq nlp n is always a nonzero integer (since p n+1 q n+1 lies strictly between the previous two convergents).

Corollary: Let a k be positive integers. Then [a 1 ;a 2 ,....] converges to an irrational α>1 .

Proof: Apply theorem to continued fraction matrices.

Example: k=1 (k 1 2 0 ) converges to an irrational number because apart from the first two matrices, we have

a k=k>2 1.5 =(b kc k) 1.5

We can develop this theory further to derive continued fraction expansions of hyperbolic trignometric functions.